Which is better, Claude Sonnet 4 or o3?

Across 50 shared benchmarks tracked on Vector Wire, Claude Sonnet 4 leads on 14 and o3 leads on 34 — o3 leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

Claude Sonnet 4 vs o3

AnthropicvsOpenAI50 shared benchmarks14–34 head-to-head

Benchmark	Claude Sonnet 4	o3
AA Agentic Index	39.2	36.1
AA Intelligence	29.8	31.1
AA-LCR	69.7	73.3
AA-Omniscience	0.2	-15.3
aider_polyglot	70.7	81.3
AIME 2024	43.4	91.6
AIME 2025	74	89.2
AIR-Bench 2024	88.3	84.5
anthropic_red_team	99.2	98.3
arc_agi_1	63.7	60.8
ARC-AGI-2	13.6	6.5
arena_vision	1207	1217
Artificial Analysis Coding Index	37.6	38.4
bbq	97.9	97.9
browsecomp	12.2	49.7
critpt	1.1	1.1
Fortress	24.4	16
frontiermath_tier_4	0	2.1
gdpval	31.2	12.8
GPQA Diamond	78	83.3
harmbench	98	98.4
HLE	10.7	20.3
IFBench	55	71.4
LiveCodeBench	68.5	84.7
livecodebench_easy	98.8	99.1
livecodebench_hard	32.6	66
livecodebench_medium	76	89.8
math_500_em	94	98.1
MMLU-Pro	84	85
MMMU	74.4	82.9
MMMU-Pro	62.4	76.4
multichallenge	57.1	60.4
MultiNRC	18.4	45.5
OmniScience Accuracy	22.7	38.6
OmniScience Non-Hallucination	70.9	12.9
scicode	40	41
simple_safety_tests	100	99
simplebench	45.5	53.1
simpleqa	15.9	49.4
swe_bench_bash	64.9	58.4
SWE-bench Verified	80.2	69.1
TAU-bench (airline)	60	52
TAU-bench (retail)	80.5	73.9
Tau2 airline	55.5	64.8
Tau2 retail	75	80.2
Terminal-Bench Hard	31.1	37.1
VisualToolBench	4.5	13.7
xstest	97.2	97.3
ZebraLogic	73.7	95.8
τ²-Bench Telecom (AA run)	64.6	80.7

Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing.